Brain Functional Network Generation Using Distribution-Regularized Adversarial Graph Autoencoder with Transformer for...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Brain functional networks offer insights into dementia disorders by revealing abnormal connectivity.
• Limited brain functional network data hinders data-driven dementia diagnosis models.
• Existing models struggle with data scarcity for robust dementia analysis.

Purpose of the Study:

• To develop a novel method for generating synthetic brain functional networks to augment datasets.
• To improve the accuracy of data-driven dementia diagnosis models using generated data.
• To enhance the robustness and stability of representation learning for brain networks.

Main Methods:

• Proposed a distribution-regularized adversarial graph auto-encoder (DAGAE) with a transformer generator.
• Utilized label distribution estimation to regularize the latent space for stable learning.
• Employed a transformer generator to create node-to-node connections, preserving topological properties.

Main Results:

• Generated brain functional networks improved prediction performance across different classifiers.
• Achieved a classification accuracy of 85.33% on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Demonstrated superior performance compared to other data augmentation models.

Conclusions:

• The proposed DAGAE model effectively generates diverse brain functional networks.
• Data augmentation with generated networks significantly improves cognitive disease diagnosis accuracy.
• The model shows promise for analyzing various cognitive disorders beyond dementia.